Artificial intelligence has become embedded across multiple stages of musculoskeletal radiology, from image acquisition to reporting, according to a 2025 scoping review published in BJR Open.
The review, led by Mickael Tordjman, MD, MS, of Icahn School of Medicine at Mount Sinai, examined current practical applications and perspectives of AI in MSK radiology, outlining the implications of 7 thematic areas: accelerated image acquisition, image interpretation, large language models (LLMs) applications, workflow integration, cost-effectiveness, liability, and education. The review showed how these advances together reshape musculoskeletal (MSK) radiology in an Artificial intelligence (AI)-enabled environment.
MSK imaging was among the earliest radiology subspecialties to explore AI, which the authors said likely reflects the broad clinical spectrum of MSK practice—including sports-related injuries, mechanical disorders, soft-tissue and bone tumors, and rheumatologic diseases, along with the large volume of imaging studies performed. Early efforts focused on fracture detection and optimization of imaging protocols. Since then, a wide range of applications have been introduced, such as deep learning (DL)-based reconstruction, automated classification of bone and soft-tissue lesions, and AI-assisted interpretation of MRI and radiographs, the authors explained.
MSK imaging encompasses examinations for back pain, mechanical disorders, sports imaging, soft-tissue and bone lesions, and rheumatologic diseases, with MRI remaining the principal advanced imaging modality.
The authors noted that DL reconstruction techniques build on parallel imaging and compressed sensing. Studies cited in the review report scan-time reductions of up to 53% without compromising diagnostic performance. This approach was first tested on the knee where DL reconstruction maintained image quality and diagnostic performance comparable to conventional MRI protocols.
Beyond speed, AI-based reconstruction has been associated with increased signal-to-noise ratio, improved fat suppression, and corrected motion and metal-related artifacts. The review notes that these improvements may enhance visualization of subtle findings, such as minor chondral defects or bone marrow edema, though aggressive under-sampling carries a risk of information loss in heavily accelerated datasets.
One of the earliest and most intuitive AI applications in MSK radiology has been fracture detection on radiographs, with multiple commercial solutions now in use, the authors noted. In settings such as emergency departments, where missed or delayed diagnoses can lead to significant morbidity, these tools have demonstrated potential clinical value.
Multiple studies have demonstrated that AI-aided radiologists performed better than AI alone or radiologists alone, the authors noted. These tools may also decrease the interpretation time and potentially reduce emergency room length of stay.
Currently at the research stage, bone tumor detection is another application of AI,that the authors noted could improve the detection and classification of malignant lesions.
Additionally, radiomics models, which analyze tumor shape, texture, and intensity patterns, have also been applied to help differentiate benign from malignant bone tumors.
The review also addressed emerging uses of LLMs in MSK imaging. Both general-purpose and medical-specific LLMs have been evaluated for tasks such as automated generation of impression sections, classification of MSK disorders, and automated generation of radiograph reports.
The authors noted that LLMs represent a new layer of AI integration, extending impact from image analysis to language-based tasks. By simplifying reports, automating classification schemes, and generating structured impressions, these models may improve communication and streamline multidisciplinary workflows. With proper safeguards, these models could become a bridge between radiologists, clinicians, and patients.
Successful real-world deployment of AI, the authors emphasized, depends on seamless integration into PACS and RIS environments, adherence to regulatory standards, and compatibility with diverse clinical practices. In MSK radiology, workflow design is essential to ensure that AI reduces reporting time and supports clinical decision-making without introducing new inefficiencies or risks and without compromising patient safety.
In addition, the authors point out that long-term adoption of AI in MSK radiology will also hinge on cost-effectiveness. The review notes that tools capable of reducing diagnostic errors, enhancing radiologist productivity, and standardizing reporting are most likely to offset implementation costs and demonstrate sustainable clinical value.
Finally, AI is best positioned as an assistive tool, reinforcing the radiologist’s expertise rather than replacing it, the authors added. They also highlighted the importance of clear guidelines, transparent performance reporting, and evolving legal frameworks to address liability risks and promote safe, responsible adoption of AI in MSK radiology.
Disclosures can be found in the study.
Source: BJR Open